Abstract.
Consider a case where cause-effect relationships between variables can be described as a directed acylic graph and the corresponding recursive factorization of a joint distribution. In order to provide the bounds on average causal effects in studies with a latent response variable, this paper proposes a graphical criterion for selecting covariates and variables caused by the response variable. The result enables us not only to judge from the graph structure whether the bounds on an average causal effect can be expressed through the observed quantities, but also to provide their closed-form expressions in case where its answer is affirmative. The graphical criterion of this paper is helpful to evaluate the bounds on average causal effects when it is difficult to observe a response variable.
Similar content being viewed by others
Acknowledgments.
The thorough and insightful comments of the reviewers and the editors on preliminary versions of this paper are gratefully acknowledged. In addition, thanks go to Zhihong Cai of Kyoto University, for her helpful comments on this paper. This research was partly supported by Grant-in-aid for Scientific Research 14780163, Ministry of Education, Culture, Sports, Science and Technology of Japan.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Kuroki, M. Bounds on average causal effects in studies with a latent response variable. Metrika 61, 63–71 (2005). https://doi.org/10.1007/s001840400324
Issue Date:
DOI: https://doi.org/10.1007/s001840400324